Heating, ventilation, and air conditioning (HVAC) is known to dominate energy consumption in commercial buildings. Increasing electricity prices worldwide is putting pressure on facility managers to reduce the energy consumption and costs associated with operating their HVAC. In this paper, we evaluate the efficacy of a data driven energy cost optimization framework for reducing HVAC cooling energy consumption via experiments conducted in a large commercial building. The major contributions of the paper are as follows. First, we develop an integer linear program (ILP) based cooling optimization framework to minimize the electricity costs incurred for cooling a building, subject to satisfying the thermal comfort of the occupants. The ILP formulation relies on data readily available from a building management system (BMS), paving the way for widespread adoption of our solution. Second, we describe the system architecture of the framework, which has been hosted on the IBM cloud platform. We outline the motivation behind implementing the solution on the cloud and highlight its key components, including the ability to use secure RESTfulAPIs alongside the ProjectHaystack open source IoT initiative to autonomously communicate with a BMS situated anywhere in the world. Third, we have deployed our framework to control theHVAC of a large office building located in northern Australia. The experiments, commenced mid Nov 2018 (and currently ongoing), carried out in two sections of the building, spanning approximately 1500m2 and housing 100 people, demonstrate that HVAC cooling energy consumption and costs can be reduced by 20%, amounting to substantial savings in annual electricity bills, without impacting the thermal comfort of the occupants.Our data driven solution is low-cost, scalable and uses sensor data commonly logged by all BMSs, providing an effective and practical mechanism for facility managers to reduce the energy consumption of their building HVAC today.